Boffin help

at 6.0 a dice and piece of paper is handy

Depends what you mean by relatively sure. Say the sample size was 1000. You'd expect 166.6 of those to be winners if the true odds are 6 for each one. If 155 were winners then there's a .16 probability that it's just down to being lucky. Now say only 145 of them were winners then the probability is .033 that you just got lucky. But if only 135 were winners then the probability would be .0036. You can work it out from excel normsdist  function I think.

I spent many hours writing a program that reads the results from Timeform which I copied and pasted into a text file. After loads of hard work I managed to create a database from Jan-July 2012. I keep intending to use more results but due to laziness I haven't done it.

I was expecting to find something that works, such as what you suggested but failed miserably.

Some months you'd be up, some down. Sometimes the difference is unbelievable!

So from my experience 7th months definitely isn't enough.

Here's some figures in a table calculated from excel:

	sample size	100	500	1000	2000	5000	10000	20000
edge(%)
1		0.482	0.460	0.444	0.421	0.376	0.327	0.159
2		0.464	0.421	0.389	0.345	0.264	0.186	0.023
3		0.447	0.382	0.336	0.274	0.171	0.090	0.001
5		0.412	0.309	0.240	0.159	0.057	0.013	0.000
7		0.377	0.242	0.161	0.081	0.013	0.001	0.000
10		0.327	0.159	0.079	0.023	0.001	0.000	0.000
15		0.251	0.067	0.017	0.001	0.000	0.000	0.000

So for odds of 6, if you have a sample size of 5000 bets and your results indicate an edge of 5% (ie a profit of 250 units) there's a probability of .057 that you don't actually have any real edge and you've just got lucky so far.

To put it another way. Say you have a million results and 16.666 % of them are wins. This means that the true odds of picking 1 winning result are 6 (1/6 * 1m). Now say you took 5000 of these 1m results at random. On average 57 times out of 1000 you can expect to get a set of results that indicate that your edge is 5% or better.

Thanks guys - had to actually read a few stats web sites just to understand your posts but much appreciated. Best of luck with your punting

No, truth is I still don't understand how you worked it out!
If sample size was 1000, (and at odds of 6to1) and I had only 155 winners instead of the expected 166.6, why is it a 16% probability that it is just luck? Sorry for being old fashioned stupid

Well you have to look at it like the average should be 166.66 out of 1000 for bf odds of 6 (ie 5/1).  But in reality you'd expect half the actual results to be under 166.66 and half of them to be over 166.66. And the distribution if it's a "normal" distribution, which I'm assuming this is, will look like a standard bell curve. The further away from 166.66 you go, the less likely you'll get that value. And the drop off in probability can be calculated using binomial distribution calculations. Try googling binomial distribution.

unfortunately you will never have enough data..

I've come to the conclusion that a system has to be much more complex than backing/laying at certain odds for it to work.

Sheppy, its just the average that comes to round 6.0. Lowest has been 1.44, highest 23.0. Up 48 points after 119 races at level stakes, so obviously interested when the data starts to become predictive, indicative, you know what I mean. Personally I'd take a 5% chance that it is just luck as being acceptable that the data I have is indicative. Im now going to google binomial distribution.

Looked at binomial distribution and have come to the conclusion I am too stupid to understand it

Daz doorstep challenge did me.

DARWIN HAD THE SAME TROUBLE BUT HE SORTED IT OUT